Hydroelectric generating set fault identification method and device, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of hydroelectric generating set fault diagnosis, and particularly discloses a hydroelectric generating set fault identification method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring multi-source measurement signal data of a to-be-measured hydroelectric generating set; the multi-source measurement signal data comprises acoustic emission signal data and vibration signal data; inputting the acoustic emission signal data and the vibration signal data into a unit fault recognition model to obtain a fault type of the to-be-detected hydroelectric generating set output by the unit fault recognition model; and the unit fault identification model is obtained by training according to the multi-source measurement signal data samples and the corresponding fault type labels, and is used for carrying out feature fusion based on extracted features of the acoustic emission signal data and the vibration signal data, and identifying the fault type of the to-be-detected hydroelectric generating unit based on the fused features. According to the invention, the fault type of the to-be-detected hydroelectric generating set can be accurately identified, the fault identification precision of the hydroelectric generating set is effectively improved, and the diagnosis performance of a hydroelectric generating set fault diagnosis system is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of hydropower unit fault diagnosis, and more specifically, relates to a hydropower unit fault identification method, device, electronic equipment and storage medium. Background Art

[0002] As an important power production equipment, the fault diagnosis technology of hydropower units is crucial to ensure the stable operation of the power system. With the development of technology, the fault diagnosis methods of hydropower units are also constantly improving. At present, there are two mainstream methods for fault diagnosis of hydropower units, namely the fault diagnosis method based on machine learning method and the fault diagnosis method based on end-to-end deep learning.

[0003] At present, the commonly used machine learning algorithms for fault diagnosis include support vector machine (SVM), k-nearest neighbor algorithm (KNN), Bayesian classification and artificial neural network (ANN). These algorithms perform well when dealing with relatively simple unit fault diagnosis tasks, but the model performance is insufficient when facing complex fault signals and noise interference. Commonly used deep learning methods for fault diagnosis include multilayer perceptron (MLP) and convolutional neural network (CNN). Although the model performance of these methods has been improved to a certain extent, most of these methods use single-channel sensor signals for diagnosis and identification. Since only part of the state monitoring data of the mechanical equipment is used, the amount of fault information contained in the data is limited, resulting in the fault identification accuracy of the diagnosis model is not high.

[0004] Therefore, how to better realize fault identification of hydropower units has become a technical problem that needs to be solved urgently in the industry. Summary of the invention

[0005] In view of the defects of the prior art, the purpose of this application is to better realize the fault identification of hydropower units, aiming to solve the problem of low accuracy of fault identification of hydropower units in the prior art.

[0006] To achieve the above objectives, in a first aspect, the present application provides a method for identifying a fault of a hydropower unit, comprising:

[0007] Acquire multi-source measurement signal data of the hydropower unit to be tested; the multi-source measurement signal data includes acoustic emission signal data and vibration signal data;

[0008] Inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model;

[0009] The unit fault identification model is trained based on multi-source measurement signal data samples and corresponding fault type labels, and is used to perform feature fusion based on the features extracted from the acoustic emission signal data and the vibration signal data, and identify the fault type of the hydropower unit to be tested based on the fused features.

[0010] Optionally, the unit fault identification model includes a feature extraction module, a feature alignment module, a feature fusion module and a fault classification module; the step of inputting the acoustic emission signal data and the vibration signal data into the unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model includes:

[0011] Inputting the acoustic emission signal data and the vibration signal data into the feature extraction module respectively, and obtaining a first feature set corresponding to the acoustic emission signal data and a second feature set corresponding to the vibration signal data output by the feature extraction module;

[0012] Inputting the first feature set and the second feature set into the feature alignment module to obtain the first feature set and the second feature set of feature alignment output by the feature alignment module; the feature alignment module is constructed based on a dual-channel convolutional neural network model;

[0013] Inputting the first feature set and the second feature set of the feature alignment into the feature fusion module to obtain the fusion feature of the acoustic vibration signal output by the feature fusion module; the feature fusion module is constructed based on the channel attention mechanism;

[0014] The acoustic vibration signal fusion feature is input into the fault classification module to obtain the fault type of the hydropower unit to be tested output by the fault classification module.

[0015] Optionally, the feature extraction module includes a first feature extraction module and a second feature extraction module;

[0016] The first feature extraction module is used for:

[0017] Perform feature calculation based on the acoustic emission signal data to determine multiple target features corresponding to the acoustic emission signal data; the multiple target features are obtained by feature analysis based on multiple time domain features, multiple frequency domain features, and multiple spectral kurtosis features of the acoustic emission signal data;

[0018] Determining a first feature set corresponding to the acoustic emission signal data according to the multiple target features;

[0019] The second feature extraction module is used for:

[0020] The vibration signal data is feature encoded using a deep autoencoder to obtain a second feature set corresponding to the vibration signal data; the deep autoencoder is trained based on vibration signal data samples.

[0021] Optionally, the deep autoencoder is an improved deep autoencoder; the improved deep autoencoder is obtained by adding a target activation function layer to the encoder in the original deep autoencoder, so as to enhance the nonlinear feature extraction capability of the encoder.

[0022] Optionally, the feature fusion module is used to:

[0023] Perform feature splicing on any acoustic emission signal feature in the first feature set of feature alignment and its corresponding vibration signal feature in the second feature set of feature alignment to obtain a plurality of spliced ​​fusion features;

[0024] Based on the channel attention mechanism, feature fusion is performed on each of the spliced ​​fusion features to obtain the acoustic vibration signal fusion feature.

[0025] Optionally, before inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model, the method further includes:

[0026] Taking the multi-source measurement signal data samples and the corresponding fault type labels as a group of training samples, obtaining multiple groups of training samples;

[0027] The multiple groups of training samples are used to train the unit fault identification model.

[0028] Optionally, the training of the unit fault identification model using the multiple groups of training samples includes:

[0029] For any set of training samples, the training samples are input into the unit fault identification model, and the prediction probability corresponding to the training samples is output;

[0030] Using a preset loss function, a loss value is calculated according to the predicted probability corresponding to the training sample and the fault type label corresponding to the training sample;

[0031] Based on the loss value, adjusting the model parameters of the unit fault identification model until the loss value is less than a preset threshold or the number of training times reaches a preset number of times;

[0032] The model parameters obtained when the loss value is less than the preset threshold or the number of training times reaches the preset number are used as the model parameters of the trained unit fault identification model to complete the training of the unit fault identification model.

[0033] In a second aspect, the present application provides a hydropower unit fault identification device, comprising:

[0034] An acquisition module, used to acquire multi-source measurement signal data of the hydropower unit to be tested; the multi-source measurement signal data includes acoustic emission signal data and vibration signal data;

[0035] An identification module, used for inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model;

[0036] The unit fault identification model is trained based on multi-source measurement signal data samples and corresponding fault type labels, and is used to perform feature fusion based on the features extracted from the acoustic emission signal data and the vibration signal data, and identify the fault type of the hydropower unit to be tested based on the fused features.

[0037] In a third aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0039] In a fifth aspect, the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0040] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the prior art:

[0041] The present application provides a method, device, electronic device and storage medium for identifying faults of hydropower units. By considering the fusion and identification of multi-source measurement signal data of the hydropower unit, a trained deep neural network unit fault identification model is used to extract features of the acquired acoustic emission signal data and vibration signal data of the hydropower unit to be tested, and the extracted acoustic emission signal features are fused with complementary features in the vibration signal. This can accurately identify the fault type of the hydropower unit to be tested, effectively improve the accuracy of hydropower unit fault identification, and enhance the diagnostic performance of the hydropower unit fault diagnosis system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of a method for identifying a hydropower unit fault provided in an embodiment of the present application;

[0043] Figure 2 is a structural schematic diagram of a unit fault identification model provided in an embodiment of the present application;

[0044] Figure 3 is a schematic diagram of the structure of an improved deep autoencoder provided in an embodiment of the present application;

[0045] Figure 4 It is a schematic diagram of the model training process of the unit fault identification model provided in the embodiment of the present application;

[0046] Figure 5 It is a structural schematic diagram of a hydropower unit fault identification device provided in an embodiment of the present application;

[0047] Figure 6 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] The terms "first", "second", etc. in the specification and claims herein are used to distinguish different objects rather than to describe a specific order of the objects. For example, a first feature set and a second feature set are used to distinguish feature sets containing different types of signal features rather than to describe a specific order of the feature sets.

[0050] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0051] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more. For example, multiple time domain features refer to two or more time domain features of the acoustic emission signal data of a hydropower unit; multiple target features refer to two or more target features, etc.

[0052] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0053] Figure 1 is a flow chart of a method for identifying a hydropower unit fault provided in an embodiment of the present application, such as Figure 1 As shown, including:

[0054] Step S1, obtaining multi-source measurement signal data of the hydropower unit to be tested; the multi-source measurement signal data includes acoustic emission signal data and vibration signal data;

[0055] Step S2, inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model;

[0056] The unit fault identification model is trained based on multi-source measurement signal data samples and corresponding fault type labels. It is used to perform feature fusion based on the features extracted from acoustic emission signal data and vibration signal data, and identify the fault type of the hydropower unit to be tested based on the fused features.

[0057] Specifically, the multi-source measurement signal data described in the embodiment of the present application refers to a data set obtained by arranging multiple types of sensors in the hydropower unit to be tested to collect sensor signals and preprocessing the collected signal data.

[0058] The acoustic emission signal data described in the embodiments of the present application refers to signal data collected by an acoustic emission sensor arranged in the hydropower unit to be tested.

[0059] The vibration signal data described in the embodiments of the present application refers to signal data collected by a vibration sensor arranged in the hydropower unit to be tested.

[0060] The unit fault identification model described in the embodiment of the present application is obtained by training based on multi-source measurement signal data samples and corresponding fault type labels, and is used to learn the intrinsic correlation between the impact of acoustic emission signal data and vibration signal data on hydropower unit faults. By performing feature fusion on the features extracted from the acoustic emission signal data and the vibration signal data, the hydropower unit fault is identified based on the fused features, thereby outputting a high-precision fault identification result.

[0061] Among them, it can be understood that the multi-source measurement signal data samples can be obtained by pre-collecting the acoustic emission signal data samples and vibration signal data samples of the hydropower unit under different operating fault conditions, and adding corresponding fault type labels to the acoustic emission signal data samples and vibration signal data samples collected under the same fault.

[0062] Here, it should be noted that the fault types of the hydropower unit may include turbine cavitation, turbine crack and other faults. Accordingly, multi-source measurement signal data samples and corresponding fault type labels may be obtained under these types of faults.

[0063] That is to say, in the embodiment of the present application, the model training samples are composed of multiple groups of multi-source measurement signal data samples carrying fault type labels.

[0064] The fault type label is predetermined according to the multi-source measurement signal data sample and corresponds to the multi-source measurement signal data sample one by one. That is, each multi-source measurement signal data sample in the training sample is predetermined to carry a corresponding fault type label.

[0065] It should also be noted that in the embodiment of the present application, the unit fault identification model can be constructed using a deep neural network model. Among them, the deep neural network model can specifically adopt a joint model including a deep convolution neural network (DCNN), a deep autoencoder (DAE) network, an attention mechanism, etc., and can also adopt a fault identification network model based on other types of neural networks.

[0066] In an embodiment of the present application, a preset deep neural network model is trained by using multi-source measurement signal data samples and corresponding fault type labels to obtain a trained unit fault identification model for use in fault identification of hydropower units.

[0067] Furthermore, in an embodiment of the present application, in step S1, signal data is collected by acoustic emission sensors and vibration sensors preset in the hydropower unit to be tested, and multi-source measurement signal data of the hydropower unit to be tested, including acoustic emission signal data and vibration signal data, is obtained through data preprocessing, including data cleaning, normalization processing and other processes.

[0068] Furthermore, in an embodiment of the present application, in step S2, the acquired acoustic emission signal data and vibration signal data of the hydropower unit to be tested are input into a pre-trained unit fault identification model. The unit fault identification model can perform feature extraction on the input acoustic emission signal data and vibration signal data, and fuse the various extracted features. Finally, the fault type of the hydropower unit to be tested is identified based on the fused features, and the fault type of the hydropower unit to be tested is output.

[0069] The method for identifying faults of hydropower units in the embodiment of the present application considers the fusion and identification of multi-source measurement signal data of the hydropower unit, uses a trained deep neural network unit fault identification model to extract features from the acquired acoustic emission signal data and vibration signal data of the hydropower unit to be tested, and uses the extracted acoustic emission signal features and complementary features in the vibration signal to fuse them, so as to accurately identify the fault type of the hydropower unit to be tested, effectively improve the accuracy of fault identification of the hydropower unit, and enhance the diagnostic performance of the fault diagnosis system of the hydropower unit.

[0070] Figure 2 is a schematic diagram of the structure of the unit fault identification model provided in the embodiment of the present application, such as Figure 2 As shown, as an optional embodiment, the unit fault identification model includes a feature extraction module 1, a feature alignment module 2, a feature fusion module 3 and a fault classification module 4; the acoustic emission signal data and the vibration signal data are input into the unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model, including:

[0071] Inputting the acoustic emission signal data and the vibration signal data into the feature extraction module 1 respectively, obtaining a first feature set corresponding to the acoustic emission signal data and a second feature set corresponding to the vibration signal data output by the feature extraction module 1;

[0072] Inputting the first feature set and the second feature set into the feature alignment module 2, obtaining the first feature set and the second feature set of feature alignment output by the feature alignment module 2; the feature alignment module is constructed based on a dual-channel convolutional neural network (PCNN) model;

[0073] The first feature set and the second feature set of feature alignment are input into the feature fusion module 3 to obtain the fusion feature of the acoustic vibration signal output by the feature fusion module 3; the feature fusion module is constructed based on the channel attention mechanism (CAM);

[0074] The acoustic-vibration signal fusion feature is input into the fault classification module 4 to obtain the fault type of the hydropower unit to be tested output by the fault classification module 4 .

[0075] Specifically, the first feature set described in the embodiment of the present application refers to a feature set obtained by performing feature extraction of different dimensions on acoustic emission signal data.

[0076] The second feature set described in the embodiment of the present application refers to a feature set obtained by performing feature extraction of different dimensions on vibration signal data.

[0077] It should be noted that in order to perform feature alignment processing on the acoustic emission signal features in the first feature set and the vibration signal in the second feature set, the number of features contained in the first feature set and the second feature set can be set to be the same, and a feature of the acoustic emission signal and a feature of the vibration signal can be randomly selected for feature alignment processing. Among them, the selection of the number of features can be set according to actual needs, and this application does not make specific restrictions on this.

[0078] The feature alignment module described in the embodiment of the present application can also be described as a "vibration signal feature alignment module", which is constructed based on the PCNN model, and can be specifically composed of two parallel CNN models with different input channel sizes and the same output channels, wherein one CNN model is used to input the first feature set, and the other CNN model is used to input the second feature set. The structure of the CNN model mainly includes a convolutional layer, an activation function, a pooling layer, and a fully connected layer.

[0079] Among them, the convolution layer is the core of the convolutional network, which is mainly used for feature extraction and noise reduction. The result of convolution is used as the feature extracted from the input data. The input of the convolution layer and the output of the previous layer are weight-shared and locally connected, so the training parameters of the model can be reduced, the learning efficiency can be accelerated, and the complexity of the network can be simplified. Since most of the data are nonlinear, if there is no activation function, it is just a simple linear combination. Therefore, introducing a nonlinear activation function after convolution can enhance the nonlinear fitting ability of the model. Specifically, the Sigmoid function can be used as the activation function.

[0080] The pooling layer follows the convolution layer, which reduces redundant feature information while achieving feature scale invariance, extracts discriminative features, and reduces the complexity of network calculations; the fully connected layer is usually located at the end of the network, integrating and transforming the features extracted by the network for final classification. Its function is to connect all nodes in the input layer to all nodes in the output layer, and map the data of the input layer to the output layer by learning weights and bias values.

[0081] The feature fusion module described in the embodiment of the present application can also be described as an "acoustic-vibration signal feature fusion module", which is mainly constructed based on a CAM network and is used to perform feature fusion processing on input acoustic emission signal features and vibration signal features.

[0082] The fault classification module described in the embodiment of the present application can be constructed using a fully connected network and a Softmax regression model, which is used to obtain a linear combination of features by fusing the features of the acoustic vibration signal through a layer of fully connected combination network, and then input into the Softmax regression model to calculate the probability of each type of fault belonging to the feature, thereby realizing the fault classification of the hydropower unit.

[0083] In an embodiment of the present application, a unit fault identification model includes a feature extraction module, a feature alignment module, a feature fusion module and a fault classification module. Thus, after the acoustic emission signal data and the vibration signal data are input into the unit fault identification model, the acoustic emission signal data and the vibration signal data are firstly input into the feature extraction module respectively, and a plurality of acoustic emission signal features and a plurality of vibration signal features of the same number are extracted respectively by using a feature extraction network corresponding to the acoustic emission signal data and a feature extraction network corresponding to the vibration signal data preset in the module, thereby obtaining a first feature set corresponding to the acoustic emission signal data and a second feature set corresponding to the vibration signal data.

[0084] Further, in an embodiment of the present application, the first feature set and the second feature set obtained above are input into a feature alignment module, and based on the feature extraction of the PCNN model, any vibration signal feature in the second feature set and its corresponding acoustic emission signal feature in the first feature set are subjected to a convolutional layer operation of the PCNN. The process can be expressed as follows:

[0085]

[0086] Where: They respectively represent the features of the output of the i-th convolution kernel in the network; Respectively represent any of the aforementioned vibration signal features and its corresponding acoustic emission signal feature; n represents the total number of features in the first feature set and the second feature set, f(·) represents the nonlinear excitation function; ω ij Represents the weight of the input signal feature and the jth convolution kernel; b jis the bias of the convolution kernel.

[0087] After convolution, the maximum pooling operation is performed:

[0088]

[0089] Where: Y j in , Represents the jth convolution kernel feature of the input feature surface, f max (·) indicates the maximum value of the function, Y i out , is the output feature of the i-th convolution kernel of the output feature surface.

[0090] In this way, the PCNN model can be used to perform deeper extraction of the vibration signal features in the second feature set and the acoustic emission signal features in the first feature set. The two signal features with different feature dimensions are extracted and transformed through two convolutional neural network models with different input channels and the same output channels, and converted into features of the same dimension. Finally, the alignment of each acoustic emission signal feature and the corresponding vibration signal feature is achieved, and the first feature set and the second feature set with feature alignment are obtained.

[0091] Furthermore, the first feature set and the second feature set of feature alignment are input into the feature fusion module, and the CAM network set inside the feature fusion module is used to perform feature fusion processing on the first feature set and the second feature set of feature alignment, and finally an acoustic-vibration signal fusion feature that fuses the acoustic emission signal feature and the vibration signal feature is obtained.

[0092] Furthermore, the acoustic-vibration signal fusion feature obtained above is input into the fault classification module, and the fault classification module predicts and calculates the acoustic-vibration signal fusion feature to determine the fault type of the hydropower unit to be tested.

[0093] The method of the embodiment of the present application constructs a unit fault identification model through network modularization, and performs feature extraction, alignment and fusion identification on acoustic emission signal data and vibration signal data in turn, which can effectively extract acoustic emission features and vibration features and realize feature fusion, effectively improving the diagnostic accuracy of the hydropower unit fault diagnosis model and solving the problem of single fault diagnosis signal and low accuracy of traditional hydropower units.

[0094] Continue to refer to Figure 2 ,Based on the content of the above embodiment, as an optional embodiment, the feature extraction module ,1 includes a first feature extraction module 11 and a second feature extraction module 12;

[0095] The first feature extraction module 11 is used to: perform feature calculation based on the acoustic emission signal data to determine multiple target features corresponding to the acoustic emission signal data; the multiple target features are obtained by feature analysis based on multiple time domain features, multiple frequency domain features and multiple spectral kurtosis features of the acoustic emission signal data;

[0096] Determine a first feature set corresponding to the acoustic emission signal data according to the plurality of target features;

[0097] The second feature extraction module 12 is used for:

[0098] The vibration signal data is feature encoded using a deep autoencoder to obtain a second feature set corresponding to the vibration signal data; the deep autoencoder is trained based on vibration signal data samples.

[0099] Specifically, the target feature described in the embodiment of the present application refers to a feature that makes the acoustic emission signal data feature more identifiable, which can be obtained by feature selection from multiple time domain features, multiple frequency domain features and multiple spectral kurtosis features of the acoustic emission signal data of the hydropower unit.

[0100] The first feature extraction module described in the embodiment of the present application refers to a module for extracting features from acoustic emission signal data, which can also be described as an "acoustic emission signal feature extraction module".

[0101] The second feature extraction module described in the embodiment of the present application refers to a module for extracting features from vibration signal data, which can also be described as a “vibration signal feature extraction module”.

[0102] Due to the huge amount of acoustic emission signal data, it is difficult to directly use deep learning methods for feature extraction. In order to extract multi-faceted features that can characterize the fault of a hydropower unit from the acoustic emission signal, in an embodiment of the present application, a multi-dimensional feature space containing time domain, frequency domain, and spectral kurtosis features is first constructed for the acoustic emission signal of the hydropower unit, and then a feature selection method based on FisherScore is used to effectively screen the features, and finally the acoustic emission signal features of the hydropower unit are extracted.

[0103] Among them, extracting the time domain features of the acoustic emission signal can effectively understand the time domain characteristics of the unit. The following Table 1 shows the 14 time domain features extracted from the acoustic emission signal of the hydropower unit, among which T1~T8 are dimensionless eigenvalues, T9~T 14 is a dimensionless eigenvalue.

[0104] Furthermore, the time domain signal can be processed into a frequency domain signal through Fourier transform, and five frequency domain features as shown in Table 2 below can be extracted. 15 ~T 17 It can describe the frequency distribution of signal components with large power spectrum components, T 18~T 19 Can be used to describe the energy concentration of the power spectrum.

[0105] Table 1

[0106]

[0107] Table 2

[0108]

[0109] In addition, spectral kurtosis can find transient components and their frequency domain locations in a signal with strong noise, and can act as a filter to recover signals submerged in a strong noise background. Suppose the short-time Fourier transform of the signal x(t) is X(t, f), and its n-order spectral moment is: S nX (f)=<|X(t,f)| n >

[0110] In the formula, <·> represents the average time operator, f and t represent frequency and time respectively. The fourth-order spectral moment after energy normalization is defined as spectral kurtosis, which can be expressed as:

[0111] In this way, the frequency band signal with large spectral kurtosis can be separated by the value of spectral kurtosis to obtain the fault signal of instantaneous impact. The commonly used characteristic values ​​of spectral kurtosis include the mean, standard deviation, skewness and kurtosis of spectral kurtosis. Therefore, the four spectral kurtosis features shown in Table 3 below can be extracted.

[0112] In an embodiment of the present application, during the unit fault identification model training stage, the Fisher Score algorithm can be used to perform feature selection on the multiple categories of features of the above-mentioned extracted acoustic emission signal data, and irrelevant or redundant features can be removed, while other original features are retained, thereby obtaining a feature subset to better describe the acoustic emission signal characteristics of the hydropower unit with minimal performance loss.

[0113] Table 3

[0114]

[0115] It should be noted that Fisher Score is a feature selection method under supervised learning, which can effectively judge sample features. The classic Fisher Score comes from Fisher's linear discriminant method, and its essence is to select sample features with small intra-class differences and large inter-class differences.

[0116] The specific implementation is as follows: Assume that a feature set {f1, f2, f3...., f m} is a training sample x selected from c (c ≥ 2) fault categories j ∈Rm , S = 1, 2, 3...N, define the training sample f i The inter-class dispersion of the features and the k-th class sample in the i-th f i The intra-class divergence S under the feature b (f i )for:

[0117]

[0118]

[0119] Combining the above two formulas, we can get the sample feature f i The Fisher Score formula is as follows:

[0120]

[0121] When the feature f i The inter-class dispersion S b (f i ) is larger, the more discrete The smaller the characteristic f i FS(f i ) value is larger, it is more conducive to feature selection.

[0122] Therefore, by performing feature selection based on the Fisher Score algorithm on the 14 time domain eigenvalues, 5 frequency domain eigenvalues, and 4 spectral kurtosis eigenvalues, the Fisher Score value of each feature is calculated. If the Fisher Score value is higher, the identifiability of the feature is better, and multiple target features with high Fisher Score values ​​are selected as the acoustic emission signal feature set.

[0123] In the embodiment of the present application, the first feature extraction module can be used to perform feature calculation on the acoustic emission signal data in a targeted manner, determine multiple target features corresponding to the acoustic emission signal data, and then group these target features into a set to obtain a first feature set corresponding to the acoustic emission signal data.

[0124] Furthermore, the DAE network in the second feature extraction module is used to perform feature encoding on the vibration signal data to obtain a second feature set corresponding to the vibration signal data.

[0125] Among them, the DAE network is trained based on vibration signal data samples. The DAE network is an unsupervised deep neural network composed of a symmetrical encoder and decoder, which can be regarded as a stack of multiple autoencoders. The encoder part of DAE consists of a series of fully connected layers with decreasing number of neurons, which is used to reduce the dimension of the original data. The output of the last layer is the reduced potential features; the decoder reconstructs the potential features into the original data. The smaller the reconstruction error, the better the feature reduction effect of DAE.

[0126] Assume that a given N-dimensional original vibration signal data sample sequence X = {x1, x2, ..., x N}, after an encoder with k fully connected layers E = [E1, E2, ..., E K ], compressed into an M-dimensional latent feature H, and passed through a decoder D = [D1, D2, ..., D K ], get the reconstructed data The reconstruction error L D The network parameters are adjusted as the loss function of the DAE network. The loss function can be expressed as:

[0127]

[0128] When L D If the value is small enough, the DAE network can effectively learn the potential features, and the DAE network training is completed.

[0129] Furthermore, in this embodiment, the encoder part of the trained DAE network can be used to perform feature encoding on the vibration signal data to complete feature extraction, and finally obtain a second feature set corresponding to the vibration signal data.

[0130] The method of the embodiment of the present application adopts a feature selection algorithm to select multiple target features that can characterize the identifiability of acoustic emission signals from multiple types of features of acoustic emission signals, so as to perform targeted feature extraction on the acoustic emission signal data of the hydropower unit, and perform feature extraction on the vibration signal data through the trained DAE network, so as to effectively obtain the feature set of the acoustic emission signal data and the feature set of the vibration signal data, effectively improve the feature extraction accuracy of the multi-source signals of the hydropower unit, and help to further improve the accuracy of subsequent fault identification of the hydropower unit.

[0131] Based on the contents of the above embodiments, as an optional embodiment, the deep autoencoder is an improved deep autoencoder; the improved deep autoencoder is obtained by adding a target activation function layer to the encoder in the original deep autoencoder, which is used to enhance the nonlinear feature extraction capability of the encoder.

[0132] Specifically, in the embodiment of the present application, in order to further explore the potential features in the original vibration signal data, inspired by the kernel function transformation, a target activation function layer can be introduced as a feature expansion layer based on the original DAE network. Among them, the target activation function layer can specifically adopt a Sigmoid activation function layer.

[0133] Figure 3 is a schematic diagram of the structure of the improved deep autoencoder provided in the embodiment of the present application, such as Figure 3 As shown, in the embodiment of the present application, the encoder E=[E1, E2, ..., E K ], including an introduced Sigmoid activation function layer and K-1 fully connected layers, where the Sigmoid activation function layer can be used as the E2 layer. In this way, the process of feature extraction of the improved DAE network can be expressed as follows:

[0134] E1=ReLU(w E1 X+b E1 );

[0135] E2=Sigmoid(w E2 E1+b E2 );

[0136] E k =ReLU(W Ek E k-1 +b Ek );

[0137] H = ReLU(w H E k +b H );

[0138] Where ReLU() and Sigmoid() represent the ReLU activation function and the Sigmoid activation function respectively; w and b represent the weight and bias of each layer respectively; E2 is the sigmoid activation function layer, which is used to use more neurons to nonlinearly map data to a high-dimensional space.

[0139] Similarly, referring to the above embodiment, the reconstruction error L D As the loss function of the improved DAE network, the network parameters are adjusted. D When the value is small enough, the improved DAE network can effectively learn the potential features. At this point, the training of the improved DAE network is completed.

[0140] Furthermore, in this embodiment, the encoder part in the trained improved DAE network can be used to perform feature encoding on the vibration signal data to complete feature extraction, and finally obtain a second feature set corresponding to the vibration signal data.

[0141] The method of the embodiment of the present application introduces a target activation function layer as a feature expansion layer on the basis of the original DAE network, and uses a fully connected layer to map the original data to a higher dimensional space, thereby decoupling the potential information in the original data, thereby enhancing the nonlinear feature extraction capability of the DAE network encoder and further improving the accuracy of feature extraction of vibration signal data.

[0142] Based on the content of the above embodiment, as an optional embodiment, the feature fusion module is used to:

[0143] Perform feature splicing on any acoustic emission signal feature in the first feature set of feature alignment and its corresponding vibration signal feature in the second feature set of feature alignment to obtain a plurality of spliced ​​fusion features;

[0144] Based on the channel attention mechanism, the features of each spliced ​​fusion are fused to obtain the fusion features of the acoustic vibration signal.

[0145] Specifically, in the embodiment of the present application, a feature fusion module based on a CAM network is used to allocate attention weights to and fuse vibration signal features and acoustic emission signal features, and finally obtain acoustic vibration signal fusion features.

[0146] More specifically, first, any acoustic emission signal feature in the first feature set of feature alignment and its corresponding vibration signal feature in the second feature set of feature alignment are spliced ​​and fused into a simple fused feature, that is, multiple spliced ​​fused features can be obtained.

[0147] Furthermore, each spliced ​​fusion feature is input into the CAM network for aggregation average pooling and maximum pooling to obtain two feature description vectors, and then the two descriptors are input into the shared network to generate channel attention features to achieve the fusion of acoustic vibration signal features, and the acoustic vibration signal fusion feature corresponding to each spliced ​​fusion feature can be obtained.

[0148] The shared network consists of an MLP network with one hidden layer. After applying MLP to each descriptor, the output features are calculated by element-wise summation. In short, the calculation of the attention mechanism can be expressed as:

[0149]

[0150] In the formula, represents element-wise multiplication, F represents input features, and F′ represents the fused features selected by channel attention correlation.

[0151] The method of the embodiment of the present application realizes the fusion of acoustic and vibration signal features by utilizing the CAM-based acoustic and vibration signal feature fusion model, and highlights the features that are more critical to fault identification and diagnosis through attention weight allocation, thereby further improving the accuracy of fault identification of hydropower units.

[0152] Based on the content of the above embodiment, as an optional embodiment, before inputting the acoustic emission signal data and the vibration signal data into the unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model, the method further includes:

[0153] Taking the multi-source measurement signal data samples and the corresponding fault type labels as a group of training samples, obtaining multiple groups of training samples;

[0154] Multiple groups of training samples are used to train the unit fault identification model.

[0155] Specifically, in an embodiment of the present invention, before inputting the acoustic emission signal data and the vibration signal data into the unit fault identification model, the unit fault identification model needs to be trained to obtain a trained unit fault identification model.

[0156] In an embodiment of the present invention, acoustic emission sensors and vibration sensors are first used to collect original acoustic emission signal data and original vibration signal data samples of hydropower units operating under different operating conditions and fault states, and classified according to different operating condition fault types, and a One-Hot encoding method is used to make a fault type label, and the collected original acoustic emission signal data and original vibration signal data are normalized, and the data are mapped to a range of 0 to 1 to accelerate the network convergence speed, and finally a data set of multi-source measurement signal data samples is formed, which specifically includes an acoustic emission signal data sample set and a corresponding vibration signal data sample set.

[0157] Therefore, the above sample data set can be divided into training set, validation set and test set in the ratio of 6:2:2. The training set is a data set used to train the deep learning model. During training, the model uses the samples in the training set to learn features and patterns in order to make predictions or classifications; the validation set is a data set used to adjust model hyperparameters and evaluate model performance. During the training process, the validation set is used to adjust the parameters of the model to prevent the model from overfitting on the training set and to help select the best model parameters; the test set is a data set used to evaluate the performance of the trained model. After the model training is completed, the test set is used to verify the model's generalization ability for unknown data.

[0158] Furthermore, in an embodiment of the present invention, the above training set sample data is used to train the unit fault identification model, and the specific training process is as follows:

[0159] The multi-source measurement signal data samples and the corresponding fault type labels are used as a group of training samples, that is, each multi-source measurement signal data sample with a fault type label is used as a group of training samples, thereby obtaining multiple groups of training samples. It can be understood that the multi-source measurement signal data samples include acoustic emission signal data samples and corresponding vibration signal data samples.

[0160] In the embodiment of the present invention, the multi-source measurement signal data samples and the fault type labels they carry are in one-to-one correspondence.

[0161] Then, after obtaining multiple sets of training samples, the multiple sets of training samples are sequentially input into the unit fault recognition model, and the unit fault recognition model is trained using the multiple sets of training samples, that is:

[0162] The multi-source measurement signal data samples in each group of training samples and the fault type labels they carry are simultaneously input into the unit fault identification model. According to each output result in the unit fault identification model, the model parameters in the unit fault identification model are adjusted by calculating the loss function value. When the preset training termination conditions are met, the entire training process of the unit fault identification model is finally completed to obtain a trained unit fault identification model.

[0163] The method of the embodiment of the present invention uses multi-source measurement signal data samples and their corresponding fault type labels as a group of training samples, and uses multiple groups of training samples to train the unit fault identification model, which is beneficial to improving the model accuracy of the trained unit fault identification model.

[0164] Based on the content of the above embodiment, as an optional embodiment, a unit fault identification model is trained using multiple groups of training samples, including:

[0165] For any set of training samples, the training samples are input into the unit fault identification model, and the prediction probability corresponding to the training samples is output;

[0166] Using the preset loss function, the loss value is calculated according to the prediction probability corresponding to the training sample and the fault type label corresponding to the training sample;

[0167] Based on the loss value, the model parameters of the unit fault identification model are adjusted until the loss value is less than a preset threshold or the number of training times reaches a preset number;

[0168] The model parameters obtained when the loss value is less than a preset threshold or the number of training times reaches a preset number are used as the model parameters of the trained unit fault identification model to complete the training of the unit fault identification model.

[0169] Specifically, the preset loss function described in the embodiment of the present invention refers to the loss function pre-set in the unit fault identification model, which is used for model evaluation; the preset threshold refers to the threshold pre-set by the model, which is used to obtain the minimum loss value and complete the model training; the preset number refers to the maximum number of model iteration training pre-set.

[0170] After obtaining multiple groups of training samples, for any group of training samples, the multi-source measurement signal data samples in each group of training samples and the fault type labels they carry are simultaneously input into the unit fault identification model, and the prediction probability corresponding to the training sample is output.

[0171] On this basis, the preset loss function is used to calculate the loss value according to the predicted probability corresponding to the training sample and the fault type label corresponding to the training sample.

[0172] Furthermore, after the loss value is calculated, the training process ends. The model parameters of the unit fault identification model are adjusted based on the loss value to update the weight parameters of each layer of the model in the unit fault identification model, and then the next training is carried out, and the model training is repeated iteratively.

[0173] During the training process, if the training results for a certain group of training samples meet the preset training termination conditions, such as the corresponding calculated loss value is less than the preset threshold, or when the current number of iterations reaches the preset number, the model loss value can be controlled within the convergence range, then the model training ends. At this time, the obtained model parameters can be used as the model parameters of the trained unit fault identification model, and the unit fault identification model training is completed, thereby obtaining a trained unit fault identification model.

[0174] The method of the embodiment of the present invention controls the loss value of the unit fault identification model within a convergence range by repeatedly iteratively training the unit fault identification model using multiple groups of training samples, thereby facilitating improving the accuracy of the unit fault identification results output by the model and improving the accuracy of hydropower unit fault identification.

[0175] Figure 4 is a schematic diagram of the model training process of the unit fault identification model provided in the embodiment of the present application, such as Figure 4 As shown, in the embodiment of the present application, the acoustic and vibration signal fault collection of the hydropower unit is first performed, and the original acoustic emission signal data and original vibration signal data of the hydropower unit under different working conditions and fault conditions are collected by using acoustic emission sensors and vibration sensors, and classified according to different working conditions, and the fault type label is made by One-Hot encoding. Then, after the data preprocessing process, a multi-source measurement signal data sample set is obtained, that is, an acoustic emission signal data sample set and a vibration signal data sample set are obtained.

[0176] Then, in order to extract multi-faceted features that can characterize the fault of hydropower unit from the acoustic emission signal, a multidimensional feature space containing time domain, frequency domain, and spectral kurtosis features is constructed for any acoustic emission signal data sample in the acoustic emission signal data sample set of the hydropower unit. Then, the target features are effectively screened based on Fisher Score, and finally the acoustic emission signal feature set of the acoustic emission signal data sample is extracted.

[0177] Next, the DAE model is trained using the vibration signal data of the hydropower unit. After extraction by the encoder and decoder of the deep autoencoder, the key information is reduced from the vibration signal, the interference and redundant components in the data are eliminated, and the DAE model parameters are continuously adjusted through the loss function until the loss value converges to obtain the trained DAE model. Then, the encoder of the trained DAE model is used to perform deep feature extraction on the vibration signal data sample corresponding to the acoustic emission signal data sample, and finally the vibration signal feature set corresponding to the vibration signal data sample is extracted.

[0178] Furthermore, the PCNN model is used to perform deeper feature extraction on any acoustic emission signal feature and the corresponding vibration signal feature of the aforementioned acoustic emission signal data sample. The two types of signal features with different feature dimensions are extracted and transformed through two parallel CNN models with different input channels and the same output channels, and converted into features of the same dimension. Finally, the alignment of the acoustic emission signal features and the corresponding vibration signal features is achieved, preparing for the subsequent acoustic vibration signal feature fusion.

[0179] Next, referring to the fusion method in the above embodiment, the vibration signal features and the corresponding acoustic emission signal features are assigned attention weights and fused using the acoustic vibration signal feature fusion model based on the CAM mechanism, and finally multiple acoustic vibration signal fusion features are obtained. The features are then classified through a series of fully connected layers and Softmax regression models, and finally the fault diagnosis results of the hydropower unit are obtained.

[0180] Furthermore, the fault diagnosis result and the prepared fault type label are used to calculate the preset loss function and adjust the parameters of the entire neural network model, thus completing a model training process. The above process is then iterated and calculated continuously until the loss value of the preset loss function is less than the preset threshold or the number of training times reaches the preset number of times, and finally the model parameters obtained when the loss value is less than the preset threshold or the number of training times reaches the preset number of times are used as the model parameters of the trained unit fault identification model to complete the training of the unit fault identification model.

[0181] The method of the embodiment of the present application uses acoustic-vibration multi-channel signals for fault diagnosis and identification, and improves the accuracy and reliability of fault diagnosis of hydropower units through the complementary features of acoustic emission signals and vibration signals; at the same time, a method of constructing a feature set and feature selection is adopted to effectively solve the problem of large processing volume of acoustic emission signals, and through Fisherscore screening, the consumption of computing resources is reduced and the efficiency of feature extraction is improved; and by using the DAE model to extract the deep features of the vibration signal, more refined and representative features can be extracted from complex data, thereby enhancing the generalization ability of the model.

[0182] The hydropower unit fault identification device provided in the present application is described below. The hydropower unit fault identification device described below and the hydropower unit fault identification method described above can be referenced to each other.

[0183] Figure 5 is a schematic diagram of the structure of a hydropower unit fault identification device provided in an embodiment of the present application, such as Figure 5 As shown, including:

[0184] An acquisition module 10 is used to acquire multi-source measurement signal data of the hydropower unit to be tested; the multi-source measurement signal data includes acoustic emission signal data and vibration signal data;

[0185] The identification module 20 is used to input the acoustic emission signal data and the vibration signal data into the unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model;

[0186] The unit fault identification model is trained based on multi-source measurement signal data samples and corresponding fault type labels. It is used to perform feature fusion based on the features extracted from acoustic emission signal data and vibration signal data, and identify the fault type of the hydropower unit to be tested based on the fused features.

[0187] It can be understood that the detailed functional implementation of each of the above-mentioned units / modules can be found in the introduction of the aforementioned method embodiment, and will not be repeated here.

[0188] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method, which will not be repeated here.

[0189] The hydropower unit fault identification device of the embodiment of the present application considers the fusion and identification of multi-source measurement signal data of the hydropower unit, uses the trained deep neural network unit fault identification model to extract features from the acquired acoustic emission signal data and vibration signal data of the hydropower unit to be tested, and uses the extracted acoustic emission signal features and the complementary features in the vibration signal to fuse them, so as to accurately identify the fault type of the hydropower unit to be tested, effectively improve the accuracy of hydropower unit fault identification, and enhance the diagnostic performance of the hydropower unit fault diagnosis system.

[0190] Based on the method in the above embodiment, the embodiment of the present application provides an electronic device, such as Figure 6 As shown, the electronic device may include: a processor (Processor) 610, a communication interface (CommunicationsInterface) 620, a memory (Memory) 630 and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the method in the above embodiment.

[0191] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0192] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0193] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0194] It is understandable that the processor in the embodiment of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0195] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0196] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0197] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0198] It should be understood that expressions such as "including" and "may include" that may be used in the present application indicate the existence of the disclosed functions, operations, or constituent elements, and do not limit one or more additional functions, operations, and constituent elements. In the present application, terms such as "including" and / or "having" may be interpreted as indicating specific characteristics, numbers, operations, constituent elements, components, or combinations thereof, but may not be interpreted as excluding the existence or possibility of adding one or more other characteristics, numbers, operations, constituent elements, components, or combinations thereof.

[0199] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for identifying faults of a hydropower unit, characterized in that: include: Acquire multi-source measurement signal data of the hydropower unit to be tested; the multi-source measurement signal data includes acoustic emission signal data and vibration signal data; Inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model; The unit fault identification model is trained based on multi-source measurement signal data samples and corresponding fault type labels, and is used to perform feature fusion based on the features extracted from the acoustic emission signal data and the vibration signal data, and identify the fault type of the hydropower unit to be tested based on the fused features.

2. The method for identifying faults of a hydroelectric generator set according to claim 1, characterized in that: The unit fault identification model includes a feature extraction module, a feature alignment module, a feature fusion module and a fault classification module; the acoustic emission signal data and the vibration signal data are input into the unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model, including: Inputting the acoustic emission signal data and the vibration signal data into the feature extraction module respectively, and obtaining a first feature set corresponding to the acoustic emission signal data and a second feature set corresponding to the vibration signal data output by the feature extraction module; Inputting the first feature set and the second feature set into the feature alignment module to obtain the first feature set and the second feature set of feature alignment output by the feature alignment module; the feature alignment module is constructed based on a dual-channel convolutional neural network model; Inputting the first feature set and the second feature set of the feature alignment into the feature fusion module to obtain the fusion feature of the acoustic vibration signal output by the feature fusion module; the feature fusion module is constructed based on the channel attention mechanism; The acoustic vibration signal fusion feature is input into the fault classification module to obtain the fault type of the hydropower unit to be tested output by the fault classification module.

3. The method for identifying faults of a hydroelectric generator set according to claim 2, characterized in that: The feature extraction module includes a first feature extraction module and a second feature extraction module; The first feature extraction module is used for: Perform feature calculation based on the acoustic emission signal data to determine multiple target features corresponding to the acoustic emission signal data; the multiple target features are obtained by feature analysis based on multiple time domain features, multiple frequency domain features, and multiple spectral kurtosis features of the acoustic emission signal data; Determining a first feature set corresponding to the acoustic emission signal data according to the multiple target features; The second feature extraction module is used for: The vibration signal data is feature encoded using a deep autoencoder to obtain a second feature set corresponding to the vibration signal data; the deep autoencoder is trained based on vibration signal data samples.

4. The method for identifying faults of a hydroelectric unit according to claim 3, characterized in that: The deep autoencoder is an improved deep autoencoder; the improved deep autoencoder is obtained by adding a target activation function layer to the encoder in the original deep autoencoder, and is used to enhance the nonlinear feature extraction capability of the encoder.

5. The method for identifying faults of a hydroelectric generator set according to claim 2, characterized in that: The feature fusion module is used for: Perform feature splicing on any acoustic emission signal feature in the first feature set of feature alignment and its corresponding vibration signal feature in the second feature set of feature alignment to obtain a plurality of spliced ​​fusion features; Based on the channel attention mechanism, feature fusion is performed on each of the spliced ​​fusion features to obtain the acoustic vibration signal fusion feature.

6. The method for identifying faults of a hydropower unit according to any one of claims 1 to 5, characterized in that: Before inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model, the method further includes: Taking the multi-source measurement signal data samples and the corresponding fault type labels as a group of training samples, obtaining multiple groups of training samples; The multiple groups of training samples are used to train the unit fault identification model.

7. The method for identifying faults of a hydroelectric generator set according to claim 6, characterized in that: The using the multiple groups of training samples to train the unit fault identification model includes: For any set of training samples, the training samples are input into the unit fault identification model, and the prediction probability corresponding to the training samples is output; Using a preset loss function, a loss value is calculated according to the predicted probability corresponding to the training sample and the fault type label corresponding to the training sample; Based on the loss value, adjusting the model parameters of the unit fault identification model until the loss value is less than a preset threshold or the number of training times reaches a preset number of times; The model parameters obtained when the loss value is less than the preset threshold or the number of training times reaches the preset number are used as the model parameters of the trained unit fault identification model to complete the training of the unit fault identification model.

8. A hydropower unit fault identification device, characterized in that: include: An acquisition module, used to acquire multi-source measurement signal data of the hydropower unit to be tested; the multi-source measurement signal data includes acoustic emission signal data and vibration signal data; An identification module, used for inputting the acoustic emission signal data and the vibration signal data into a unit fault identification model to obtain the fault type of the hydropower unit to be tested output by the unit fault identification model; The unit fault identification model is trained based on multi-source measurement signal data samples and corresponding fault type labels, and is used to perform feature fusion based on the features extracted from the acoustic emission signal data and the vibration signal data, and identify the fault type of the hydropower unit to be tested based on the fused features.

9. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program runs on a processor, the processor is caused to execute the method according to any one of claims 1 to 7.

Images